Authors
Shihan Zhao, Jianru Zhang, Yanan Wu, Linlin Li, Siyuan Shen, Xingjun Zhu, Guoyan Zheng, Jiahua Jiang, Wuwei Ren
Published in
IEEE transactions on medical imaging. Volume PP. Aug 26, 2026. Epub Aug 26, 2026.
Abstract
Fluorescence Molecular Tomography is a promising technique for non-invasive 3D visualization of fluorescent probes, but its reconstruction remains challenging due to the inherent ill-posedness and the necessity of adopting presumed tissue optical properties, whose exact ground-truth values are arduous to acquire. Emerging deep learning approaches have shown promising reconstruction results with higher spatial resolution, but mostly rely on supervised training with a large scenario-specific paired dataset and often degrade under optical-parameter mismatch. To address these problems, we propose μNeuFMT, a self-supervised neural-field reconstruction framework that integrates a differentiable finite-element-method solver in the loss function. A core advantage of μNeuFMT lies in its dual capability: it reconstructs fluorophore spatial distributions while adaptively optimizing paired background optical parameters, thereby lowering the dependence on prior information regarding background absorption and scattering coefficients. Extensive numerical, phantom, and in vivo lymph node and tumor imaging validation studies demonstrate consistent gains in accuracy and robustness across varying acquisition settings, even with moderately biased optical parameter initializations. We envision μNeuFMT as a meaningful step toward more reliable fluorescence molecular imaging in complex biological tissues. Our code and example data are publicly available at: https://github.com/Star6azeR/NeuFMT.
PMID:
42647692
Bibliographic data and abstract were imported from PubMed on 27 Aug 2026.
Read full publication at:
Please sign in
to see all details.
Advertisement
Stats
- Recommendations n/a n/a positive of 0 vote(s)
- Views 7
- Comments 0